A case for reframing automated medical image classification as segmentation